CausalInferenceChallenges/Propensity_Models.py at master · behavioral-data/CausalInferenceChallenges · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Thu Jun 20 17:20:41 2019

@author: peterawest
"""

import numpy as np
from sklearn.linear_model import Perceptron, SGDClassifier, LogisticRegressionCV
from sklearn.metrics import log_loss
from sklearn.calibration import CalibratedClassifierCV

from sklearn.model_selection import PredefinedSplit


# assuming binary treatment
def CE(P, Z):
    return -np.sum(Z*np.log(P))/len(P)

# assuming binary treatment
def accuracy(P, Z):
    return 1.*(np.sum((P > 0.5)*Z) + np.sum((P <= 0.5)*(1-Z))) / len(P)


def train_propensity_model(propensity_model, dataset, data_test=False):

    ## first do training and validation


    propensity_model.fit(dataset)
    out_dict = {}
    out_dict['learning_curve'] = propensity_model.learning_curve


    X = [data[0] for data in dataset.train_epoch(true_set = True)]
    Z = [data[1] for data in dataset.train_epoch(true_set = True)]
    Y = [data[2] for data in dataset.train_epoch(true_set = True)]

    # get training score
    P_train = propensity_model.score(X)
    out_dict['train_acc'] = accuracy(P_train, np.array(Z))
    out_dict['train_CE'] = CE(P_train, np.array(Z))
    out_dict['P_train'] = P_train
    out_dict['Z_train'] = Z
    out_dict['Y_train'] = Y


    X = [data[0] for data in dataset.valid_epoch()]
    Z = [data[1] for data in dataset.valid_epoch()]
    Y = [data[2] for data in dataset.valid_epoch()]

    # get validation score
    P_val = propensity_model.score(X)
    out_dict['val_acc'] = accuracy(P_val, np.array(Z))
    out_dict['val_CE'] = CE(P_val, np.array(Z))
    out_dict['P_val'] = P_val
    out_dict['Z_val'] = Z
    out_dict['Y_val'] = Y

    ## if we have test data, also return test scores
    if data_test:
        X = [data[0] for data in dataset.test_epoch()]
        Z = [data[1] for data in dataset.test_epoch()]
        Y = [data[2] for data in dataset.test_epoch()]

        P_test = propensity_model.score(X)

        out_dict['test_acc'] = accuracy(P_test, np.array(Z))
        out_dict['test_CE'] = CE(P_test, np.array(Z))
        out_dict['P_test'] = P_test
        out_dict['Z_test'] = Z
        out_dict['Y_test'] = Y


    return propensity_model, out_dict

def train_PW_model(propensity_model, dataset, data_test=False):

    ## first do training and validation


    propensity_model.fit(dataset)
    out_dict = {}
    out_dict['learning_curve'] = propensity_model.learning_curve


    X = [data[0] for data in dataset.train_epoch(true_set = True)]
    C = [data[1] for data in dataset.train_epoch(true_set = True)]
    Y = [data[2] for data in dataset.train_epoch(true_set = True)]

    # get training score
    P_train = propensity_model.score(X)
    out_dict['train_acc'] = accuracy(P_train, np.array(C))
    out_dict['train_CE'] = CE(P_train, np.array(C))
    out_dict['P_train'] = P_train


    X_true = [(x[0], x[2]) for x in X]
    out_dict['P_train_true'] = propensity_model.score(X_true)
#    out_dict['Z_PW_train'] = Z_PW
    #out_dict['Y_train'] = Y


    X = [data[0] for data in dataset.valid_epoch()]
    C = [data[1] for data in dataset.valid_epoch()]
    Y = [data[2] for data in dataset.valid_epoch()]

    # get training score
    P_val = propensity_model.score(X)
    out_dict['val_acc'] = accuracy(P_val, np.array(C))
    out_dict['val_CE'] = CE(P_val, np.array(C))
    out_dict['P_val'] = P_val
#    out_dict['Z_PW_val'] = Z_PW
    X_true = [(x[0], x[2]) for x in X]
    out_dict['P_val_true'] = propensity_model.score(X_true)

    ## if we have test data, also return test scores
    if data_test:
        X = [data[0] for data in dataset.test_epoch()]
        C = [data[1] for data in dataset.test_epoch()]
        Y = [data[2] for data in dataset.test_epoch()]

        # get training score
        P_test = propensity_model.score(X)
        out_dict['test_acc'] = accuracy(P_test, np.array(C))
        out_dict['test_CE'] = CE(P_test, np.array(C))
        out_dict['P_test'] = P_test
#        out_dict['Z_PW_test'] = Z_PW
        X_true = [(x[0], x[2]) for x in X]
        out_dict['P_test_true'] = propensity_model.score(X_true)


    return propensity_model, out_dict


class Logreg_Propensity_Model:
    def __init__(self, n_train, n_val, penalty='l2', scoring='neg_log_loss', class_weight='balanced'):

        self.n_train = n_train
        self.n_val = n_val

        # define validation split for sklearn
        q = np.zeros(n_train + n_val)
        q[n_train:] = -1

        ps = PredefinedSplit(q)

        self.model =  LogisticRegressionCV(Cs=20,
                        cv = ps,
                       random_state=0,
                       penalty = penalty,
                       scoring = scoring,
                       class_weight=class_weight,
                      refit = False)


        pass

    def fit(self, X_train, Z_train, X_val, Z_val):

        # define full data to be used
        # THIS SHOULD MATCH n_train, n_val in definition
        X = X_train + X_val
        Z = Z_train + Z_val
        X = np.array(X)
        Z = np.array(Z)

        # these must match for validation to work
        assert(len(X_train) == self.n_train)
        assert(len(X_val) == self.n_val)

        self.model.fit(X,Z)


    def score(self, X):
        X = np.array(X)

        return self.model.predict_proba(X)[:,1]